ESG as Priced Crash Insurance: State-Dependent Tail Risk and Deconfounding Evidence

Battle Count: 82

Highly relevant for quantitative risk management and portfolio construction. The state-dependent crash insurance framework directly informs tail-risk hedging strategies, ESG tilting in factor models, and regime-aware position sizing. The DML deconfounding approach is applicable to any high-dimensional asset pricing problem where selection bias and multicollinearity are concerns. The finding that ESG protection is concentrated in extreme left tails (1st-2nd percentiles) during stress months is actionable for tail-risk overlay strategies. However, the paper is more academic/causal-inference focused than directly implementable as a trading signal.

The multi-stage framework requires: (1) constructing a market stress indicator from panel data; (2) estimating regime-specific logit models with clustered standard errors; (3) running conditional quantile regressions with stratified month-block bootstrap (800 replicates); and (4) implementing Double Machine Learning with cross-fitting, Neyman-orthogonal scores, and flexible nuisance estimators (Lasso, RF, GBM). The DML component is the most complex, requiring careful sample splitting, cross-fitting, and residualization. The overall pipeline is non-trivial but well-specified in the paper. Requires proficiency in causal inference, quantile regression, and ML-based econometrics.

The paper uses publicly available data sources (MSCI ESG ratings, Compustat via WRDS, S&P 500 constituents). Methodology is well-documented with explicit equations for stress indicator, crash logits, quantile regressions, and DML procedure. However, no code repository is mentioned, and MSCI ESG data requires a subscription. The drawdown-based truncation rule and bootstrap procedures are clearly specified. Month-block bootstrap with 800 replicates is described.